#!/bin/bash

#用途：按下数字查看显卡水位并告警，
#Author:rohing
#History:20231214 Version:1.0
# 设置警告阈值
THRESHOLD=70

#
CURRENT_TIME=$(date +"%Y-%m-%d %H:%M:%S")
# 获取主机名和IP
HOSTNAME=$(hostname)
#这是内网地址
IP=$(hostname -I)
#这是公网地址
IP2=$(curl -s ifconfig.me)


check_nvidia=$(nvidia-smi |awk 'NR==10'| awk -F ' ' '{printf "%.0f", $9 /$11*100 }')
check1_nvidia(){
if [  "$check_nvidia" -ge "$THRESHOLD"  ]; then
     #   # 构建payload
     PAYLOAD=$(cat <<-EOF
{

"msgtype": "markdown",
"markdown": {
"title":"Linux服务器显存告警",
"text":"
##### Linux服务器显存告警 \n
>  ##### <font color=#67C23A> 【服务器: </font> <font color=#FF0000> $HOSTNAME</font>】 的显存空间超过阀值</font> :<font color=#FF0000> $THRESHOLD%  </font> \n
>  ##### <font color=#67C23A> 【服务器IP】</font> :<font color=#FF0000> 内网地址: $IP 公网IP地址: $IP2 </font> \n
>  ##### <font color=#67C23A> 【告警时间】</font> :<font color=#FF0000> $CURRENT_TIME </font> \n
>  ##### <font color=#67C23A> 【Linux服务器显存】:已使用 <font color=#FF0000>$check_nvidia%</font> \n
"
}
}
EOF
    )
    # 发送告警 (自行替换钉钉Webhook机器人的URL)
    curl -H "Content-Type: application/json" -X POST -d "$PAYLOAD"  https://oapi.dingtalk.com/robot/send?access_token=7ef7db65342f433a321b353453453453453453453535341505952d5ba13248
    echo $PAYLOAD
  fi
  }
check1_nvidia